15 Time Space

Stat405
Visualising time & space

Hadley Wickham
Wednesday, 21 October 2009

1. New data: baby names by state
2. Visualise time (done!)
3. Visualise time conditional on space
4. Visualise space
5. Visualise space conditional on time


Baby names by state
Top 100 male and female baby names
for each state, 1960–2008.
480,000 records (100 * 50 * 2 * 48)
Slightly different variables: state, year,
name, sex and number.
To keep the data manageable, we will
look at the top 25 names of all time.

CC BY http://www.ﬂickr.com/photos/the_light_show/2586781132

Subset
Easier to compare states if we have
proportions. To calculate proportions,
need births. But could only ﬁnd data from
1981.
Then selected 30 names that occurred
fairly frequently, and had interesting
patterns.


Aaron Alex Allison Alyssa Angela Ashley
Carlos Chelsea Christian Eric Evan
Gabriel Jacob Jared Jennifer Jonathan
Juan Katherine Kelsey Kevin Matthew
Michelle Natalie Nicholas Noah Rebecca
Sara Sarah Taylor Thomas


Getting started

library(ggplot2)
library(plyr)

bnames <- read.csv("interesting-names.csv",
stringsAsFactors = F)

matthew <- subset(bnames, name == "Matthew")


Time | Space


0.04

0.03
prop

0.02

0.01

1985 1990 1995 2000 2005
year

0.04

0.03
prop

0.02

0.01

1985 1990 1995 2000 2005
qplot(year, prop, data = matthew,year
geom = "line", group = state)

AK AL AR AZ CA CO CT DC
0.04
0.03
0.02
0.01
DE FL GA HI IA ID IL IN
0.04
0.03
0.02
0.01
KS KY LA MA MD ME MI MN
0.04
0.03
0.02
0.01
MO MS MT NC NE NH NJ NM
0.04
prop

0.03
0.02
0.01
NV NY OH OK OR PA RI SC
0.04
0.03
0.02
0.01
SD TN TX UT VA VT WA WI
0.04
0.03
0.02
0.01
WV WY
0.04
0.03
0.02
0.01
198519952005 19902000 198519952005 19902000 198519952005 19902000 198519952005 19902000
19902000 198519952005 19902000 198519952005 19902000 198519952005 19902000 198519952005
year

AK AL AR AZ CA CO CT DC
0.04
0.03
0.02
0.01
DE FL GA HI IA ID IL IN
0.04
0.03
0.02
0.01
KS KY LA MA MD ME MI MN
0.04
0.03
0.02
0.01
MO MS MT NC NE NH NJ NM
0.04
prop

0.03
0.02
0.01
NV NY OH OK OR PA RI SC
0.04
0.03
0.02
0.01
SD TN TX UT VA VT WA WI
0.04
0.03
0.02
0.01
WV WY
0.04
0.03
0.02
0.01
198519952005 19902000 198519952005 19902000 198519952005 19902000 198519952005 19902000
19902000 198519952005 19902000 198519952005 19902000 198519952005 19902000 198519952005

last_plot() + facet_wrap(~ state)year

Your turn

Pick some names out of the list and
explore. What do you see?
Can you write a function that plots the
trend for a given name?


show_name <- function(name) {
name <- bnames[bnames$name == name, ]
qplot(year, prop, data = name, geom = "line",
group = state)
}

show_name("Jessica")
show_name("Aaron")
show_name("Juan") + facet_wrap(~ state)


0.04

0.03
prop

0.02

0.01

qplot(year, prop, data = matthew, geom1995 "line", 2000
1985 1990 = group = state) +
2005
geom_smooth(aes(group = 1), se year size = 3)
= F,

0.04

0.03
prop

0.02

0.01

So we only get one smooth
for the whole dataset
qplot(year, prop, data = matthew, geom1995 "line", 2000
1985 1990 = group = state) +
2005
geom_smooth(aes(group = 1), se year size = 3)
= F,

Two useful tools
Smoothing: can be easier to perceive
overall trend by smoothing individual
functions
Indexing: remove initial differences by
indexing to the ﬁrst value.
Not that useful here, but good tools to
have in your toolbox.


library(mgcv)
smooth <- function(y, x) {
as.numeric(predict(gam(y ~ s(x))))
}

matthew <- ddply(matthew, "state", transform,
prop_s = smooth(prop, year))

qplot(year, prop_s, data = matthew, geom = "line",
group = state)


index <- function(y, x) {
y / y[order(x)[1]]
}

matthew <- ddply(matthew, "state", transform,
prop_i = index(prop, year),
prop_si = index(prop_s, year))

qplot(year, prop_i, data = matthew, geom = "line",
group = state)
qplot(year, prop_si, data = matthew, geom = "line",
group = state)


Your turn
Create a plot to show all names
simultaneously. Does smoothing every
name in every state make it easier to see
patterns?
Hint: run the following R code on the next
slide to eliminate names with less than 10
years of data


longterm <- ddply(bnames, c("name", "state"),
function(df) {
if (nrow(df) > 10) {
df
}
})


qplot(year, prop, data = bnames, geom = "line",
group = state, alpha = I(1 / 4)) +
facet_wrap(~ name)

longterm <- ddply(longterm, c("name", "state"),
transform, prop_s = smooth(prop, year))

qplot(year, prop_s, data = longterm, geom = "line",
group = state, alpha = I(1 / 4)) +
facet_wrap(~ name)
last_plot() + facet_wrap(scales = "free_y")


Space


Spatial plots

Choropleth map:
map colour of areas to value.
Proportional symbol map:
map size of symbols to value


juan2000 <- subset(bnames, name == "Juan" & year == 2000)

# Turn map data into normal data frame
library(maps)
states <- map_data("state")
states$state <- state.abb[match(states$region,
tolower(state.name))]

# Merge and then restore original order
choropleth <- merge(juan2000, states,
by = "state", all.y = T)
choropleth <- choropleth[order(choropleth$order), ]

# Plot with polygons
qplot(long, lat, data = choropleth, geom = "polygon",
fill = prop, group = group)


45

40
prop
0.004
0.006
lat

0.008
35
0.01

30

−120 −110 −100 −90 −80 −70
long

What’s the problem
with this map?
45 How could we ﬁx it?

40
prop
0.004
0.006
lat

0.008
35
0.01

30

−120 −110 −100 −90 −80 −70
long

ggplot(choropleth, aes(long, lat, group = group)) +
geom_polygon(fill = "white", colour = "grey50") +
geom_polygon(aes(fill = prop))


Problems?

What are the problems with this sort of
plot?
Take one minute to brainstorm some
possible issues.


Problems
Big areas most striking. But in the US (as
with most countries) big areas tend to
least populated. Most populated areas
tend to be small and dense - e.g. the East
coast.
(Another computational problem: need to
push around a lot of data to create these
plots)


●

45
●

●

●
●

40 ●
●

● prop
● ●
● 0.004
● ● 0.006
lat

● ●
● 0.008
35
● ● ● 0.010

●

●
30

●

−120 −110 −100 −90 −80 −70
long

mid_range <- function(x) mean(range(x))
centres <- ddply(states, c("state"), summarise,
lat = mid_range(lat), long = mid_range(long))

bubble <- merge(juan2000, centres, by = "state")
qplot(long, lat, data = bubble,
size = prop, colour = prop)

ggplot(bubble, aes(long, lat)) +
geom_polygon(aes(group = group), data = states,
fill = NA, colour = "grey50") +
geom_point(aes(size = prop, colour = prop))


Your turn

Replicate either a choropleth or a
proportional symbol map with the name
of your choice.


Space | Time


Your turn

Try and create this plot yourself. What is
the main difference between this plot and
the previous?


juan <- subset(bnames, name == "Juan")
bubble <- merge(juan, centres, by = "state")

ggplot(bubble, aes(long, lat)) +
geom_polygon(aes(group = group), data = states,
fill = NA, colour = "grey50") +
geom_point(aes(size = prop, colour = prop)) +
facet_wrap(~ year)


Aside: geographic data

Boundaries for most countries available
from: http://gadm.org
To use with ggplot2, use the fortify
function to convert to usual data frame.
Will also need to install the sp package.


# install.packages("sp")

library(sp)
load(url("http://gadm.org/data/rda/CHE_adm1.RData"))

head(as.data.frame(gadm))
ch <- fortify(gadm, region = "ID_1")
str(ch)

qplot(long, lat, group = group, data = ch,
geom = "polygon", colour = I("white"))


This work is licensed under the Creative
Commons Attribution-Noncommercial 3.0 United
States License. To view a copy of this license,
visit http://creativecommons.org/licenses/by-nc/
3.0/us/ or send a letter to Creative Commons,
171 Second Street, Suite 300, San Francisco,
California, 94105, USA.


15 Time Space

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Mehr von Hadley Wickham

Mehr von Hadley Wickham (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

15 Time Space